Using Deep Learning to Support Clinical Decision-Making: The Case of Alzheimer’s Disease Diagnosis

Abstract

Alzheimer’s disease is a chronic, progressive brain disorder that leads to a gradual decline in memory and cognitive functions. In this study, N-VGG16, an advanced deep learning model, is proposed. The model builds upon the VGG16 architecture, incorporating key enhancements to improve its ability to classify neurodegenerative conditions. The model processes structural neuroimaging data using a refined pipeline that applies adaptive histogram equalization for image enhancement and employs data augmentation techniques to address class imbalance issues. A major contribution of this work is the use of gradient-based localization, which allows the model’s predictions to be linked to specific brain regions affected by the disease. Evaluation using a standardized dataset showed that the model achieved a high classification accuracy of 99.69%, successfully distinguishing between different clinical stages of Alzheimer’s disease. Furthermore, visual interpretation confirmed that the model consistently focused on brain areas commonly associated with the disease. These findings highlight the model’s potential to support clinical decision-making by offering both accurate diagnoses and interpretable insights.